As I was studying refraction, I began wondering what path would light take when entering a non-homogeneous transparent medium, i.e. a certain material in which the refraction index $n$ varies (smoothly) through space: $n=n(x,y,z)$.

My guess is that the path chosen (let's forget Feynman QED for the time being) could be the one which minimizes travel time, so here's the mathematical question:

Q: Given a differentialble non-negative scalar field $\phi:\mathbb{R}^2\longrightarrow\mathbb{R}^+$ and a smooth (or at least $\mathcal{C}^1$) curve $\gamma:[0,1]\longrightarrow\mathbb{R}^2$ such that $\gamma(0)=P$ and $\gamma(1)=Q$ with $P,Q\in\mathbb{R}^2$, how to determine the time it takes to travel from $P$ to $Q$ assuming that in every point $\gamma(\lambda)=(x(\lambda),y(\lambda))$ the modulus of the speed is $\phi(x(\lambda),y(\lambda))$ and that the direction of travel never changes (i.e. we go directly from $P$ to $Q$ without stopping or going back for blueberries)?

In other words I impose the condition that \begin{equation} \|\frac{d\gamma(\lambda(t))}{dt}\|=\phi(\gamma(\lambda)) \end{equation}

I am confused by the dependency of $\gamma$ on time, and am unsure about what parametrisation of $\gamma$ should fit the case. Should it be arc length? Does it even matter? I cannot think clearly for some reason. Thank you!


You just need to invoke Fermat's principle: light travels the trajectory which minimizes the total time. Let $\tau$ be the total time spent by the light when it goes from $P$ to $Q$ under the curve $\gamma(s) = (q^1(s),q^2(s))$, then:

$ \tau = \int_P^Q dt = \int_P^Q \frac{dl}{v} = \frac{1}{c}\int_P^Q n(\mathbf{r})dl = \int_P^Q n(\mathbf{r(s)})\sqrt{g_{ij}\dot q^i(s)\dot q^j(s)}ds, $

where $c$ is the speed of light in the vacuum, $v$ is the speed of the light in a media with refractive index $n(\mathbf{r})$, $l$ is a physical length and $s$ a parameter. You can take $s$ to be arc length, but it would be harder to solve the minimization problem since we must do it while constraining $\|\gamma'(s)\|=1$. Just use the Euler-Lagrange equations for the lagrangian

$ L(q^i,\dot q^i) = n(\mathbf{r})\sqrt{g_{ij}\dot q^i\dot q^j}. $

By solving the Euler-Lagrange equations for the generalized coordinates $q^i$, you obtain the curve $\gamma$ that describes the trajectory of the light in the non-homogeneous media (which is your primary concerning). Moreover, you can calculate the total time $\tau$ by resolving the above integral along the curve you just find. At the end, just replace $\|\gamma'(s)\|$ by $\phi(\gamma(s))$.

About the parameter, remember that the length of a curve (and, therefore, the total time) doesn't depend on the specific parameter choice. Just pay attention on the domain of it (knowing its geometrical/physical meaning usually helps).


I guess you are trying to construct a functional of the curve $\gamma$ and then minimize the functional to obtain the equation of motion for $\gamma(t)$?

Let $(x,y)$ be the coordinates on the plane and $v(x,y)$ be the speed on each point. I would parametrize the curve by time $t$ and then write down the time of travel by $$ \int v^{-1}(x,y) dl$$ where $ dl^2 = dx^2 +dy^2$ , is the infinitesimal length traveled. Thus, along the trajectory $\gamma(t)=(x(t),y(t))$ $$dl = \sqrt{\dot{x}^2+\dot{y}^2}dt $$ The functional we have is $$ \int f[x,y,\dot{x},\dot{y}]dt, \quad f= \frac{\sqrt{\dot{x}^2+\dot{y}^2}}{v(x,y)}.$$ By calculus of variation or directly the Euler-Lagrange equation, you can get the ODEs for $(x(t),y(t))$. After solving the equations of motion under your boundary condition, you can find the time used.

  • $\begingroup$ Thank you mastrok, I was definitely going to consider the variational approach but the question is slightly different- namely I consider a particular curve $\gamma$ (not necessarily the optimal curve) and would like to calculate the travel time. $\endgroup$ – marco trevi Feb 27 '15 at 9:55
  • $\begingroup$ oh I am sorry, if the curve is given then $t$ is just calculated by that integral. $\endgroup$ – mastrok Feb 27 '15 at 9:58
  • $\begingroup$ but houw do you find the parametrization of $\gamma$ with time if you have only the support of the curve? $\endgroup$ – marco trevi Feb 27 '15 at 9:59
  • $\begingroup$ How is the curve given? Are $x(t), y(t)$ given or they are given under some parametrization which is not necessarily $t$? $x(\lambda), y(\lambda)$ in which the time dependence of $\lambda(t)$ is unknown? $\endgroup$ – mastrok Feb 27 '15 at 10:02
  • $\begingroup$ Actually I am realizing that it does not matter since arc length does not depend upon the particular parametrization... $\endgroup$ – marco trevi Feb 27 '15 at 10:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.